Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
The Real Options to Shutdown, Startup, and Abandon: Empirical
Evidence
Stein-Erik Fleten, Erik Haugom, and Carl J. Ullrich∗
June 5, 2012
ABSTRACT
The purpose of this paper is to examine empirically the real options to shutdown, startup,
and abandon. The exercise is made possible by a dataset with detailed information for 1,121
individual plants for the period 2001–2009, a total of 8,189plant-year observations. To the best
of our knowledge, the data are unique in scope and level of detail. We find strong evidence of
real options effects for shutdown and abandonment decisions. We find that uncertainty about
the outcome of ongoing deregulation in retail electricity markets significantly decreases the
likelihood of shutting down operating plants.
JEL CLASSIFICATION CODES:
KEY WORDS: Real options, uncertainty, compound exchange options, spark spread, commodity prices.
∗Fleten is at the Norwegian University of Science and Technology, NO–7491, Trondheim, Norway. E-mail: [email protected]. Haugom is at Lillehammer University College and Norwegian University of Science and Technology,NO-7491, Trondheim, Norway. E-mail: [email protected] is at the Department of Finance, Insurance, and BusinessLaw, Pamplin College of Business, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061. Tel: (540) 231-2343, and email: [email protected]. The authors gratefully acknowledge Jonas Andersson, Paul D. Clark II, Gunnar Eskeland,RamanKumar, Kristin Linnerud, Steve Marshall, Johannes Mauritzen, Peter Molnar, Julio Riutort, the participants at the 2011 ElCarbon-Risk seminar at the Norwegian University of Science and Technology, and participants at the 2012 Midwest Finance Associationmeetings for useful comments and suggestions.
I. Introduction
The theory of real options predicts that, in the face of irreversible switching costs and uncertainty in cash
flows, major changes in assets are subject to hysteresis, andcan be structured as options. For example,
the opportunity to invest in, shutdown (or mothball), restart, or abandon a production asset can be cast as
call and put options on the present value of the cash flows of the asset. While real options theory is well
established and accepted, empirical tests of the theory remain scarce, owing mainly to a lack of data.
The purpose of this paper is to shed light on the empirical drivers of the exercise decisions for the
real options to shutdown, startup, and abandon (which we refer to collectively as status changes) existing
production assets. We conduct our tests using a dataset withdetailed information for 1,121 individual
power plants. To the best of our knowledge, the data are unique in scope and level of detail.
Our results show that future profitability is the single mostimportant determinant of status changes.
Higher future profitability leads to fewer shutdowns and abandonments, and more startups. For plants
which were previously shutdown, the size of the plant also isan important determinant of status changes.
Larger plants are less likely to be abandoned, and more likely to be restarted.
Our contribution includes providing empirical evidence for the theory of real options. Consistent with
the theory, we show that increases in profit margin volatility decrease the probability of shutting down an
operating plant. Increases in profit margin volatility alsodecrease the probability of abandoning a plant
which previously was shutdown, but this effect is limited tosmall plants. Our results regarding real options
effects in startup decisions are inconclusive.
We find that regulatory uncertainty is a key driver for shutdown decisions, but only marginally impor-
tant for startup and abandonment decisions. During times ofregulatory uncertainty plants are less likely to
be shutdown.
1
The remainder of the article is structured as follows. Section II provides a review of existing literature,
Section III gives an overview of the data, and in Sections IV and V we present the empirical results. Section
VI concludes.
II. Literature Review
Researchers have developed option valuation techniques toderive predictions of investment behavior of the
firm. Examples include Brennan and Schwartz (1985), Majd andPindyck (1987), McDonald and Siegel
(1985), and McDonald and Siegel (1986). In these models, increases in variability in the market value of
the completed project provide incentive to delay investment.
Empirical studies on real options include Quigg (1993), whouses data on land transactions to show that
a real options model has some explanatory power for market prices, over and above net present value. Bu-
lan, Mayer, and Somerville (2009) investigate condominiumdevelopment and find that increased volatility
reduces probability of investment, and that a real options model explains the data better than a model of
risk aversion. Kellogg (2010) finds that Texas oil companiesreduce their drilling activity when volatility
rises, and that the magnitude of this change is consistent with real options theory.
A. Shutdown, Startup, and Abandonment
McDonald and Siegel (1985) point out that the owner of an operating project holds the option to shut the
project down. Brennan and Schwartz (1985) specialize a realoptions model for the case of a commodity
mine. The paper most similar to ours is Moel and Tufano (2002), who evaluate empirically the predictions
of the Brennan and Schwartz (1985) model by examining the shutdown and startup decisions for 285 gold
mine properties for the 1988-1997 time period. They find thata real options model describes well the
empirical data.
2
Our work differs from that of Moel and Tufano (2002) in important ways. First, we focus directly on the
drivers of status changes. Second, we include a measure of regulatory uncertainty. Third, we also examine
the option to abandon a plant, an alternative which Moel and Tufano (2002) do not consider. Finally, our
dataset has approximately four times as many observations as the data used by Moel and Tufano (2002).
III. Data
In this section we describe the data in detail. The primary data sources are the Energy Information Ad-
ministration, NYMEX, and wholesale electricity market system operators. Interest rate data come from
the U.S. Federal Reserve Bank. Table I presents summary statistics for the plant-specific variables in our
sample, while Table II presents summary statistics for macroeconomic and real options variables.
A. Plant Data
The main data source for this paper is Form 860 collected and disseminated by the Energy Information
Administration (hereafter EIA), the statistical arm of theU.S. Department of Energy. Form 860 contains
detailed data for nearly every power plant in the United States, both existing and planned. We consider
plants from three major wholesale electricity markets - Pennsylvania-New Jersey-Maryland (PJM), the
New England Independent System Operator (ISO-NE), and the New York Independent System Operator
(NYISO) - for the 2001-2009 time period.2 The choice of areas and sample period is driven by (i) the
availability of electricity price data and (ii) significantchanges in Form 860 beginning in 2001.
We focus on “peaking” plants as these should be more subject to the factors expected to influence
shutdown, startup, and abandonment decisions.3 The final data set contains 8,189 plant-year observations
2Specifically, we include plants located in Connecticut, Delaware, Illinois, Indiana, Kentucky, Maine, Maryland, Mas-sachusetts, Michigan, New Hampshire, New Jersey, New York,North Carolina, Ohio, Pennsylvania, Rhode Island, Tennessee,Vermont, Virginia, Washington D.C., and West Virginia.
3We retain only simple cycle combustion turbines (CT). The fuel type is either low sulfur fuel oil (DFO), i.e., EIA fuel typesDFO, FO1, FO2, or FO4, or natural gas (NG). Baseload technologies, such a coal-fired and nuclear plants, operate more-or-lesscontinuously for the duration of their useful lives. Also, fuel prices for baseload technologies are very low and stable.
3
on 1,121 individual plants. Summary statistics for the age (to the nearest year), size (in megawatts, MW),
and efficiency (in %) are presented in Table I.4 Summary statistics are presented separately for regulated
and non-regulated plants.5
Status
For our purposes, the key variable from EIA Form 860 is the “status” of the plant. The relevant status codes
are
• OP - operating,
• BU - backup,
• SB - standby, and,
• RE - retired.
Because of changing definitions, and because there are relatively few plants classified as BU, we choose to
combine status codes BU and SB into a single category SB, standby. Details can be found in Appendix A.
4Ages are calculated based upon the first year a plant appears in the database. The efficiency reported in the tables and usedinthe regressions is literallyenergy out
energy in. See the spark spread discussion below.5The EIA defines a regulated entity as follows. See http://www.eia.gov/tools/glossary/index.cfm.
For the purpose of EIA’s data collection efforts, entities that either provide electricity within a designated franchisedservice area and/or file forms listed in the Code of Federal Regulations, Title 18, part 141 are considered regulatedentities. This includes investor-owned electric utilities that are subject to rate regulation, municipal utilities,federaland state power authorities, and rural electric cooperatives. Facilities that qualify as cogenerators or small powerproducers under the Public Utility Regulatory Power Act (PURPA) are not considered regulated entities.
4
B. Supply and Demand Data
Because electricity cannot be stored, available supply (capacity) must always exceed contemporaneous
demand in order to prevent blackouts. We measure supply adequacy by reserve margin. Reserve margin
for regionk and yeart is defined to be
RMk,t = (Ck,t −Dk,t)/Dk,t , (1)
whereCk,t is the yeart capacity in regionk andDk,t is the yeart demand in regionk, both measured in
MW. The raw data come NERC’s 2009 Electricity Supply and Demand (ES&D) database.
Projected reserve margin serves as our proxy for future profitability. Lack of storability impies that,
when demand approaches available supply, electricity prices increase at an increasing rate.6 The lower is
the reserve margin, the less excess capacity there is in the system, and the higher are wholesale electricity
prices. Thus projected reserve margin acts as an (inverse) proxy for the future profitability of the plant.
Low reserve margins imply high future profitability and viseversa.
Timing Issues
Form 860 must be filed by mid-February each year. We take the data reported, for example, in the 2005
Form 860 to be effective as of the end of calendar year 2004. Inthe regressions that follow we use only
those data which were available as of the end of 2004 in order to predict a status change (shutdown, startup,
or abandonment) during the next year, i.e., by the end of 2005. Any such change would show up in the
2006 Form 860.
At the time the 2005 Form 860 was filed, the 2004 ES&D database was most recent available. The
2004 ES&D database contains actual supply and demand data for 2003 and projections for 2004-2013. In
6See, for example, Mount, Ning, and Cai (2006) and Ullrich (2008).
5
trying to predict whether a plant has a status change in 2005,we use the projected 2005 reserve margin
from the 2004 ES&D database.
C. Plant Efficiency
The efficiency of a power plant is measured by itsheat rate. The heat rate of planti, HRi, is the amount
of fuel, measured in millions of British thermal units (MMBtu), required to generate one unit of electricity,
measured in megawatt hours (MWh). A lower number indicates greater efficiency. We use two sources for
heat rate data. Our primary source is the CEMS (Continuous Emissions Monitoring Systems) data from
the EPA.7 CEMS data is available for 631 of the 1,121 plants in our sample. Heat rate data were included
in Form 860 for 1990-1995. These data are available for 312 plants for which no CEMS data is available.
Heat rates for the remaining 178 plants are estimated based on the age and size of the plant. Details are in
Appendix B.
For ease of interpretation, we convert heat rates into energy conversion efficiencies. Heat rates have
units of MMBtuMWh . BothMMBtuandMWhmeasure energy. The two are related by a scale factor. In particular,
there are 3.41275MMBtu in one MWh. We can thus convert a plant’s heat rate into a dimensionless
conversion efficiency as
EFFi =3.41275
HRi∗100% (2)
whereEFFi is the conversion efficiency of planti which has heat rateHRi. For example, a plant with a
heat rate of 10MMBtu/MWhhas a conversion efficiency of 34.1%.
7See http://camddataandmaps.epa.gov/gdm/index.cfm?fuseaction=prepackaged.select.
6
D. Shutdown and Startup Costs
According to the theory, important determinants of the decision to shutdown and/or startup a plant are the
cost involved doing so, both the one time costs and continuing costs. The following discussion is based
upon phone interviews with industry experts.8 Details are in Appendix C.
In general, the cost to shutdown a plant is small relative to the cost to restart a plant. The cost to
restart varies with the level of maintenance performed while the plant is shutdown, and therefore is a
function of managerial priorities. In the regressions which follow we ignore the cost to shutdown and we
focus on one single technology (simple cycle combustion turbines), thereby eliminating variation across
technology types. We proxy for startup costs by calculatingthe amount of time (in years) that a plant has
been shutdown. The assumption is that a plant which has been shutdown for a long period of time has
higher startup costs than an otherwise similar plant which has been shutdown for a shorter length of time.
E. Spark Spread Volatility
The difference between the market values of electricity andfuel is referred to as thespark spread. Consider
plant i which has heat rateHRi, burns fuel j, and is located in regionk. We calculate the plant-specific
spark spread, or profit margin, expressed in units of dollarsper megawatt hour ($/MWh), for dayn as
SPRDi jk,n = Peleck,n −HRi ∗Pf uel
j,n −VOMi, (3)
wherePeleck,n is the dayn electricity price ($/MWh) in regionk, Pf uel
j,n is the dayn fuel price ($/MMBtu) for
fuel j, andVOMi is the variable operation and maintenance cost ($/MWh) for plant i. Daily spot prices
8In particular we thank Steve Marshall of Lakeland Electric and Paul D. Clark II of the City of Tallahassee for sharing theirinsights and experience.
7
for New York Harbor No. 2 Oil and NYMEX Henry Hub natural gas are taken from the EIA website.
Electricity prices come from the PJM, ISO-NE, and NYISO websites.9
Spark spread volatility is then the standard deviation of the daily spark spread over yeart.
SPRDSDi jk,t = STDEVTn=1
(
SPRDi jk,n)
, (4)
whereT is the number of days in yeart.
A plant comprises a series of call options written on the spark spread.10 Spark spread volatility thus
has two effects on the decision to make a status change. First, increased uncertainty increases the value of
waiting for more information as in a traditional real options framework. Second, because a plant is itself an
option on the spark spread, increased volatility increasesthe option value of the plant. Both of these effects
imply that higher spark spread volatility should reduce theprobability of shutting down and abandoning
plants. For startup decisions, these two effects act in opposite directions. While increased uncertainty
increases the value of waiting, higher spark spread volatility also increases the value of the underlying
asset, i.e., the plant in question.
F. Regulatory Uncertainty
Before the advent of retail competition in the U.S., customers located in a particular utility’s service terri-
tory were captive customers of that utility. The utility wasrequired to maintain enough resources to meet
9Consistent with our focus on peaking plants, we use electricity prices for the peak period of the day, defined to be the 16 hourperiod from hour ending 7 through hour ending 22. We obtain daily peak prices by taking the simple average of the hourly spotprices during the peak period.
10Because a plant itself can be considered an option, the option to, e.g., startup a plant which has previously been shutdownhas as the underlying asset an exchange option. Such an option is referred to as a compound exchange option and is examinedin detail in Carr (1988). Deng (2000) develops a regime switching model with jumps to fit spot electricity prices, subsequentlyuses the model to value plants, and compares the model valueswith actual sales prices of recently sold plants. Drawing oninfor-mation available in financial futures prices for electricity and fuels, Deng, Johnson, and Sogomonian (2001) describe valuationof generation and transmission assets. Tseng and Barz (2002) use Monte Carlo simulations and dynamic programming valuepower plants. Thompson, Davison, and Rasmussen (2004) model hydroelectric and thermal power stations as contingent claims,and use a numerical approach to optimize the operation. In these papers, the parameters of energy price dynamics, together withpower plant characteristics, determine plant values and real option exercise rules. In contrast, this paper analyzes empirically theshutdown, startup, and abandonment decisions of existing plants.
8
the demand of its captive customers. Deregulation of retailelectricity markets allows customers to choose
electricity suppliers. The prospect of retail competitionleft utilities in the position of possibly losing (or
gaining) a significant portion of native demand. If the utility’s neighbors have lower cost generation avail-
able, and the utility loses some of its native demand when retail competition is implemented, then a plant
which was economic when used to meet native demand in the regulated world might not be needed un-
der retail competition. Also, retail competition might mean that a plant which would not have run under
regulation will be profitable again.
Deregulation of retail electricity markets in the U.S. is taking place at the state level. The EIA pub-
lishes a descriptive summary of state-level deregulation.This information, supplemented by state utility
commission information, allows the construction of a state-level retail competition index.11 The index is a
discrete variable taking on values from 1 to 5, which correspond to:
1. no activity,
2. investigation underway,
3. competition recommended,
4. law passed requiring retail competition, and,
5. competition implemented.
The index measures competition in the retail market. When the index takes a value of two (2), there
is uncertainty about whether the state will implement retail competition. When the index takes a value of
three (3), there is uncertainty about the form retail competition will ultimately take. We define a regulatory
uncertainty indicator variable (REGUNCERT) which takes a value of one when the competition index is
equal to two or three, and which takes a value of zero otherwise. Consistent with real options theory, we
expect firms to be less likely to make changes in the status of existing generators when there is uncertainty
about the outcome of retail deregulation.
11A similar index was developed independently by Delmas and Tokat (2005).
9
IV. Shutdown
Consider a plant which is operating (status OP) in the current year. Next year, the plant may either continue
to operate (remain in status OP) or move to standby (SB). We define a “shutdown” to be movement from
status OP in yeart to status SB in yeart +1.12 Table III documents the occurence of shutdowns by year.
For example, of the 832 plants which were operating in 2004, 820 continued to operate in 2005 while 12
were shutdown. For the full sample there are a total of 76 instances of shutdown versus 6,539 instances of
a operating plant remaining in operating mode.
Table IV presents comparative univariate statistics for plants which were shutdown and those which
continued to operate. The descriptive variables are divided into three categories - macroeconomic, plant-
specific, and real options, i.e., measures of uncertainty.
Beginning with the macro variables, plants tend to be shutdown when projected reserve margins are
high. Higher reserve margins imply low future profitability. Plants are more likely to be shutdown when
future profitability is low.
Slower economic growth means slower growth in the demand forelectricity and therefore higher re-
serve margins. Slower economic growth also tends to reduce interest rates. Reserve margin and interest
rates are negatively correlated13 and plants tend to shutdown when interest rates are lower.
However, theeffectof low interest rates is to increase the present value of future cash flows. Hence we
would expect interest rates to have a positive relationshipwith shutdowns. The lower (higher) are interest
rates, the lower (higher) should be the probability that a plant will shutdown. The univariate statistics in
Table IV suggest exactly the opposite - interest rates tend to be lower when plants shutdown. We believe
that, when considered in isolation, interest rates are simply proxying for reserve margin. The multivariate
12While it is possible to move directly from status OP (operating) to status RE (retired), such moves are rare and are not drivenpurely by spark spread economics.
13In our data the simple correlation coefficient between interest rates and reserve margin is -0.35. In PJM, where the majorityof status changes take place, the correlation is -0.60.
10
analysis below confirms this conjecture. When we control forreserve margin, the effect of interest rates on
shutdown probabilities is positive.
Regulatory status is an indicator variable which takes a value of zero for regulated plants and one for
non-regulated plants. Only 22.4% of the shutdown events involve plants which are regulated, while 54.4%
of the instances of a plant continuing to operate involve regulated plants. Plants which shutdown are on
average older, less efficient, and smaller than plants whichcontinue to operate. All of these differences are
significant at the 5% or 1% level.
Spark spread standard deviation and the regulatory uncertainty indicator variable are both measures of
uncertainty and ought to matter if real options effects are important. Consistent with real options theory,
the table shows that shutdowns are less likely when (i) sparkspread volatility is higher, and, (ii) there is
more uncertainty about the outcome of retail deregulation.
A. Binary Logit Regression
Consider planti which burns fuelj and is located in regionk. We begin our multivariate analysis using a
binary logit specification, as follows.14
ISBi,t+1 = α+(β1∗RMk,t+1)+ (β2∗T10t)+ (β3∗REGSTi)+ (β4∗EFFi)+ (β5∗SIZEi)
+(β6∗SPRDSDi jk,t)+ (β7∗REGUNCERTt)+ ε, (5)
where
ISBi,t+1 is an indicator variable which takes the value of zero if plant i was operating in yeart and operating
in yeart +1, and which takes a value of one if planti was operating in yeart and shutdown in year
t +1,
14We do not include the age of the plant as a regressor. Because older plants tend to be less efficient, age and efficiency arehighly collinear.
11
RMk,t+1 is the projected reserve margin for regionk and yeart +1,
T10t is the ten year treasury rate for yeart,
REGSTi is an indicator variable which take the value of zero if planti is non-regulated and one if planti
is regulated,
EFFi is the efficiency of planti,
SIZEi is the capacity of planti,
SPRDSDi jk,t is the standard deviation of yeart spark spread for planti which burns fuelj and is located
in regionk, and,
REGUNCERTt is an indicator variable which takes a value of one in years inwhich the outcome of retail
deregulation is uncertain and a value of zero otherwise.
The first five regressors,RM throughSIZEshould matter in both a traditional discounted cash flow analysis
and a real options framework. The last two regressors,SPRDSDandREGUNCERT, are measures of un-
certainty and should matter if the owners of plants considerreal options effects when making shutdown de-
cisions. Table V presents the results. The table presents the average marginal effects(
∂Prob(ISB= 1)/∂x)
of each independent (x) variable. For the indicator variables (REGSTand REGUNCERT) the table
presents the change in the probability of a shutdown when thevariable changes from zero to one. We
begin by including each independent variable separately. Each coefficient is highly significant and the
signs are consistent with the summary statistics in Table IV.
Analyzing each variable separately allows us to get a feel for which of the variables is most important.
Future profitability has the most explanatory power for the shutdown decision. The coefficient onRM
is positive indicating that plants are more likely to be shutdown when there is a greater excess of exist-
ing capacity. Higher reserve margins imply lower wholesaleelectricity prices and therefore less valuable
plants. Among the individual regressions, theRM regression has the greatest psuedo-R2, the greatest log-
likelihood, and the lowest values for both information criteria statistics,AIC andBIC. The two next best
12
individual regressors are the real options variablesSPRDSDandREGUNCERT, indicating that owners of
power plants do consider uncertainty when making the decision to shutdown a plant.
The last column of Table V shows that, with one exception, theinsights gained from the individual
regressions continue to hold when all the independent variables are included in the same regression. The
exception is that the sign ofT10 changes from negative to positive, consistent with our priors about the
effect of interest rates on the option to shutdown. Most importantly, both uncertainty variables remain
significant.
Consistent with real options theory, uncertainty reduces the likelihood of shutting down an operating
plant. The coefficients on bothSPRDSDandREGUNCERTare negative and significant. Plants for which
the spark spread was more volatile during the previous year are less likely to be shutdown. The more
uncertainty there is about retail deregulation, the less likely is a plant to be shutdown.
Figure 1 plots the probability of shutdown as a function of reserve margin, based on the regression re-
sults from Table V. The top panel presents the probability ofshutting down an operating plant as a function
for reserve margin for the cases of regulatory uncertainty (blue circles) and no uncertainty (red squares).
The figure makes clear the importance of future profitability. At low values of reserve margin (high fu-
ture profitability), the probability of shutting down an operating plant is small regardless of the regulatory
environment. At higher values of reserve margin (lower values of future profitability) the probability of
shutting down an operating plant increases dramatically, but only for the case in which there is no regula-
tory uncertainty. With regulatory uncertainty the probability of shutting down an operating plant is small,
regardless of reserve margin.
The bottom panel of Figure 1 presents the probability of shutting down an operating plant as a function
of reserve margin for three values of spark spread volatility - $10/MWh (blue circles), $30/MWh (red
squares), and $100/MWh (green triangles).15 When reserve margin is low (future profitability is high), the
15From Table II the minimum and mean values for spark spread volatility in our sample are $12/MWh and $31/MWh, respec-tively. While the maximum spark spread volatility exceeds $180/MWh, this observation is a significant outlier. (The value of
13
probability of shutting down an operating plant is small, irrespective of spark spread volatility. In this case
the spark spread options which comprise the plant are effectively in-the-money and optionality does not
constitute a large part of the plant’s value. At high levels of spark spread volatility, the plant is out-of-the-
money and optionality is the main source of the plant’s value. Therefore spark spread volatility is important
at high reserve margins.
V. Startup and Abandonment
Consider a plant which is on standby (SB) in the current year.Next year the plant may either startup
(move to status OP), remain on standby (SB), or be abandoned (move to status RE). Table VI documents
occurrences of these alternatives by year. For example, of the 188 plants which were on standby in 2004,
153 were still on standby in 2005, 22 were started up, and 13 were abandoned. For the entire sample, there
are a total of 184 instances of startup and 78 instances of abandonment.
We proceed as follows. First we consider binary logit regressions for startup and abandonment in
isolation. Then we use a multinomial logit regression to consider the startup and abandonment decisions
simultaneously.
A. Startup
Table VII presents comparative univariate statistics for plants which were started up and those which re-
mained shutdown. Plants tend to startup when projected reserve margins are low, i.e., when future prof-
itability is high. Consistent with the discussion above, weexpect startups to be more likely when interest
rates are low. Table VII shows exactly the opposite - startups tend to happen when interest rates are high
again reflecting the negative correlation between interestrates and reserve margin.
spark spread volatility at the 99th percentile is $69.22/MWh.) We chose to use $10/MWh, $30/MWH, and $100/MWh in Figure1 to approximately represent the minimum, mean, and maximumvalues observed in our sample.
14
As is the case for shutdowns, startups are more likely when the owner of the plant in question is non-
regulated. Plants which startup are on average younger, more efficient, and larger. Plants which startup
have been shutdown for a shorter period of time (1.16 years) that plants which remain on standby (2.55
years). Recall that“Time Shutdown”is our proxy for startup costs. Plants which startup have lower startup
costs than plants which remain on standby.
Turning to the real options variables, Table VII shows that startups tend to occur when uncertainty
about the outcome of retail deregulation is low. However, the table also shows that plants which startup
have higher spark spread volatility than plants which remain shutdown. We address this issue further in the
discussion of the regression results below.
Startup Binary Logit Regression
We test for drivers of startups using a binary logit specification similar to that from the previous section.
IOPi,t+1 = α+(β1∗RMk,t+1)+ (β2∗T10t)+ (β3∗REGSTi)+ (β4∗EFFi)+ (β5∗SIZEi)
+(β6∗SBTIMEi,t)+ (β7∗SPRDSDi jk,t)+ (β8∗REGUNCERTt)+ ε, (6)
where
IOPi,t+1 is an indicator variable which takes the value of zero if generator i was on standby in yeart and on
standby in yeart + 1, and which takes a value of one if generatori was on standby in yeart and
operating in yeart +1,
SBTIMEi,t is the length of time, in years, that planti has been shutdown as of yeart,
and all the other variables are as defined above. Table VIII presents the results. The table presents the
average marginal effects(
∂Prob(IOP = 1)/∂x)
of each independent (x) variable. For the indicator variables
15
(REGSTandREGUNCERT) the table presents the change in the probability of a startup when the variable
changes from zero to one.
For startup decisions, the single best predictive variableis the time the plant has been shutdown,
SBTIME, for which the coefficient is negative and highly significant. The longer a plant has been shut-
down, the more it costs to restart and the less likely it is to startup.
The last column of Table VIII presents the results for the full model. The key drivers of the startup
decision areRM, SIZE, andSBTIME. Startups are more likely when future profitability is higher, for
larger plants, and when startup costs are lower. Regulatoryuncertainty reduces the probability of starting
up a plant which was previously shutdown, but this effect is only marginally significant.
In the individual regression the coefficient onSPRDSDis positive and highly significant. Taken on its
face, this result indicates that plants are more likely to startup when uncertainty is higher. In the overall
regression, the coefficient on spark spread volatility is reduced in magnitude from the individual regressions
and is no longer significant. We view this result as inconclusive and not as evidence against the predictions
of real options theory. On the one hand, higher spark spread volatility increases the value of the option to
wait for more information and therefore reduces the probability of startup. However, higher spark spread
volatility also increases the option value of the plant itself and therefore increases the probability of startup.
Our regression is unable to differentiate between these effects.
Figure 2 plots the probability of startup as a function of reserve margin, based on the regression re-
sults from Table VIII. The top panel presents the probability of starting up a plant which was previously
shutdown as a function for reserve margin for the cases of regulatory uncertainty (blue circles) and no
uncertainty (red squares). The probability of starting up aplant which was previously shutdown is a de-
creasing function of reserve margin (increasing function of future profitability). At higher values of reserve
margin (lower values of future profitability) the probability of starting up a plant which was previously
shutdown decreases. The magnitude of the decrease is greater for the case in which there is no regulatory
uncertainty.
16
The bottom panel of Figure 2 presents the probability of starting up a plant which was previously
shutdown as a function for reserve margin for three values ofspark spread volatility - $10/MWh (blue
circles), $30/MWh (red squares), and $100/MWh (green triangles). Again we see that the probability of
starting up a plant which was previously shutdown is greaterwhen reserve margin is low (future profitability
is high), irrespective of spark spread volatility. As reserve margin increases (future profitability decreases),
the probability of starting up a plant which was previously shutdown declines. Consistent with the results
in Table VIII, the probability of startup does not vary much for different values spark spread volatility.
B. Abandonment
Table IX presents comparative statistics for plants which were abandoned (or retired, RE) and those which
remained on standby (SB). Plants tend to be abandoned when projected reserve margins are high, i.e., when
future profitability is low. Plants which were abandoned were on average older, less efficient, and much
smaller than plants which were not abandoned. Abandonmentstook place when spark spread volatility was
low and when uncertainty about retail deregulation was low.
Abandonment Binary Logit Regression
We test for drivers of abandonments using a binary logit specification similar to that from the previous
sections.
IREi,t+1 = α+(β1∗RMk,t+1)+ (β2∗T10t)+ (β3∗REGSTi)+ (β4∗EFFi)+ (β5∗SIZEi)
+(β6∗SBTIMEi,t)+ (β7∗SPRDSDi jk,t)+ (β8∗REGUNCERTt)+ ε, (7)
where
IREi,t+1 is an indicator which is equal to zero if planti was on standby both in yeart and in yeart +1, and
equal to one if planti was on standby in yeart and retired in yeart +1,
17
and all the other variables are as defined above. Results are given in Table X. The table presents the
average marginal effects(
∂Prob(IRE = 1)/∂x)
of each independent (x) variable. For the indicator variables
(REGSTandREGUNCERT) the table presents the change in the probability of an abandonment when the
variable changes from zero to one.
The coefficient on reserve margin,RM, is positive and highly significant. As with the shutdown results,
future profitability is the single most important factor is making the abandonment decision. Plants are more
likely to be abandoned when their future values are low. Interest rates, plant efficiency, and plant size also
impact the decision to abandon a plant. Plants are more likely to be abandoned when interest rates are high
and the present value of future cash flows is low. Less efficient plants and smaller plants also are more
likely to be abandoned.
The coefficient on spark spread volatility,SPRDSD, is negative and significant. Higher spark spread
volatility (i) increases the value of waiting for more information, and (ii) increases the option value of the
plant in question. Both of these effects reduce the likelihood of abandonment. This result indicates that
plant owners are aware of the option value of existing plants. In the full model, regulatory uncertainty is
not important for making the abandonment decision. Becauseplants which were previously shutdown are
“out-of-the-game” already, the prospect of losing customers with the advent of retail competition is less
important for abandonment decisions.
Figure 3 plots the probability of abandonment as a function of reserve margin, based on the regression
results from Table X. The top panel presents the probabilityof abandoning a plant which was previously
shutdown as a function for reserve margin for the cases of regulatory uncertainty (blue circles) and no
uncertainty (red squares). At low values of reserve margin (high future profitability), the probability of
abandoning a plant which was previously shutdown is small regardless of the regulatory environment. At
higher values of reserve margin (lower values of future profitability) the probability of abandoning a plant
which was previously shutdown increases dramatically withor without regulatory uncertainty.
18
The bottom panel of Figure 3 presents the probability of abandoning a plant which was previously shut-
down as a function for reserve margin for three values of spark spread volatility - $10/MWh (blue circles),
$30/MWh (red squares), and $100/MWh (green triangles). As reserve margin increases (future profitabil-
ity decreases), the probability of shutting down an operating plant depends on spark spread volatility. The
probability of abandonment is much greater at low values of spark spread volatility, i.e., when the option
value of the plant is low.
C. Startup and Abandonment Multinomial Logit Regression
Consider the following multinomial regression specification.
IOPREi,t+1 = α+(β1∗RMk,t+1)+ (β2∗T10t)+ (β3∗REGSTi)+ (β4∗EFFi)+ (β5∗SIZEi)
+(β6∗SBTIMEi,t)+ (β7∗VALUESDi jk,t)+ (β8∗REGUNCERTt)+ ε, (8)
where
IOPREi,t+1 is an indicator which is equal to zero if planti was on standby in yeart and operating in yeart +1,
equal to one if planti was on standby both in yeart and in yeart +1, equal to two if planti was on
standby in yeart and retired in yeart +1
and all the other variables are as defined above. Results are given in Table XI. The table presents the
average marginal effects(
∂Prob(IRE = 1)/∂x)
of each independent (x) variable. For the indicator variables
(REGSTandREGUNCERT) the table presents the change in the probability of an abandonment when the
variable changes from zero to one.
The coefficients are substantially similar to those found inthe binary logit analyses from the previous
subsections. Unlike the binary startup regression, the coefficient on regulatory uncertainty is not significant
at the 10% level.16 Because these plants are not operating, they are not producing revenue and changes in
16The coefficient onREGUNCERTis significant at the 11% level for both startup and abandonment.
19
the regulatory environment will not decrease their contribution to the owner’s profits. As with the binary
logit results, spark spread uncertainty matters for the abandonment decision.
The advantage of a multinomial logit regression is that it allows us to consider the startup and aban-
donment decisions simultaneously. Figure 4 plots, on the same graph, the probabilities of startup (OP,
red squares), standby (SB, blue circles), and abandonment (RE, green triangles) for a plant which was
previously shutdown as a function of reserve margin. The upper panel presents the cases of regulatory
uncertainty (right) and no uncertainty (left). At low reserve margins startup is relatively more likely and
abandonment is relatively less likely. As reserve margin increases the probability of startup decreases while
the probability of abandonment increases. The probabilityof remaining on standby forms a “frown” with
the highest probability of remaining on standby occurring at intermediate levels of reserve margin. The
figure shows that while regulatory uncertainty does increase the probability of remaining on standby and
decrease the probabilities of startup and abandonment, theeffect is small and not statistically significant.
The lower panel of Figure 4 presents the cases of low ($10/MWh, left) and high ($100/MWh, right)
spark spread volatility. Comparison of the lower left and lower right panels shows that the probability of
startup is relatively insensitive to spark spread volatility. However, the probability of abandonment varies
greatly with spark spread uncertainty. Because plants which are on standby tend to be less efficient than
plants which are operating, more of the (potential) plant value derives from optionality. As a result, the
probability of abandoning a plant is highly sensitive to thelevel of spark spread volatility.
D. Size Matters for Abandonment
In all of the regressions (binary and multinomial) for startup and abandonment, the coefficient onSIZE
is significant. The average size for plants which startup is 46.6 MW (see Table VII), while the average
size for plants which are abandoned is 11.9 MW (see Table IX).Figure 5 contains histograms plotting
the distribution of size for plants which startup (top panel) and which are abandoned (bottom panel). The
figure makes obvious that all of the plants in our sample whichare abandoned are small, less than 25 MW.
20
Figure 6 plots the probabilities of of startup (OP, red squares), standby (SB, blue circles), and abandon-
ment (RE, green triangles) as a function of reserve margin for the cases of low ($10/MWh, left) and high
($100/MWh, right) spark spread volatility. The top panel presents the probabilities for large plants (>25
MW) and the bottom panel presents the probabilities for small plants (<25 MW). Focusing first on large
plants, the low and high volatility plots are similar. Sparkspread volatility does not matter much for large
plants.
Turing to small plants, the probability of abandonment is very sensitive to spark spread volatility. At
high values of spark spread volatility, the probability of abandonment remains low. But, at low values of
spark spread volatility, the probability of abandonment increases significantly as reserve margin increases.
The main conclusion to come from Figure 6 is that, in our sample at least, the real options effects of
spark spread volatility are limited small plants. It may be that large plants were moved to standby to work
on specific issues. Perhaps the owners of these plants planned to shut them down for some sort of extended
maintenance and then restart the plants.17 That is, perhaps the large plants which were shutdown were
never candidates for abandonment.
VI. Conclusions
We examine the real options to shutdown, startup, and abandon existing power plants. The single most
important driver of status changes in future profitability.We find strong evidence of real options effects
for shutdown and abandonment decisions. In particular, we find that increases in profit margin uncertainty
reduce the probability that an operating plant will be shutdown. Increases in profit margin uncertainty also
reduce the probability that a plant which was previously shutdown will be abandoned, but this effect seems
to be limited to small plants.
17This information is not available in the EIA 860 filing.
21
While we find that profit margin uncertainty has no effect on the decision to startup a plant which was
previously shutdown, we believe this result to be inconclusive. Because the plant itself is an option, higher
uncertainty has two opposing effects on startup decisions.While higher uncertainty increases the value
of the option to wait for more information and therefore reduces the probability of startup, higher profit
margin uncertainty also increases the value of the plant itself, increasing the probability of startup. We are
unable to separate these effects in the present paper.
Uncertainty about the outcome of deregulation in retail markets reduces the likelihood of shutting
down plants which are operating. Plants which are profitablein a captive market may not be profitable
when retail customers are allowed to choose their electricity suppliers. Our evidence suggests that owners
of such plants delay the shutdown decision until the outcomeof retail deregulation is known.
We find only weak evidence that regulatory uncertainty impacts the decision to either startup or aban-
don a plant which was previously shutdown. Plants which are shutdown are “out-of-the-game” already and
therefore not producing profit for the owner. The outcome of the deregulation process in not as important
for these plants.
References
Brennan, M. J., and E. S. Schwartz, 1985, “Evaluating natural resource investments,”Journal of Business,
58, 135–157.
Bulan, L., C. Mayer, and C. T. Somerville, 2009, “Irreversible investment, real options, and competition:
Evidence from real estate development,”Journal of Urban Economics, 65, 237–251.
Carr, P., 1988, “The Valuation of Sequential Exchange Opportunities,” The Journal of Finance, 43(5),
1235–1256.
Delmas, M., and Y. Tokat, 2005, “Deregulation, governance structures, and efficiency: the U.S. electric
utility sector,”Strategic Management Journal, 26(5), 441–460.
22
Deng, S.-J., 2000, “Stochastic models of energy commodity prices and their applications: Mean-reversion
with jumps and spikes,” working paper, Georgia Institute ofTechnology.
Deng, S.-J., B. Johnson, and A. Sogomonian, 2001, “Exotic electricity options and the valuation of elec-
tricity generation and transmission assets,”Decision Support Systems, 30, 383–392.
Kellogg, R., 2010, “The Effect of Uncertainty on Investment: Evidence from Texas Oil Drilling,” working
paper 16541, National Bureau of Economic Research.
Majd, S., and R. S. Pindyck, 1987, “Time to build, option value, and investment decisions,”Journal of
Financial Economics, 18, 7–27.
McDonald, R. L., and D. R. Siegel, 1985, “Investment and the valuation of firms when there is an option
to shut down,”International Economic Review, 26, 331–349.
, 1986, “The value of waiting to invest,”Quarterly Journal of Economics, 101, 707–728.
Moel, A., and P. Tufano, 2002, “When Are Real Options Exercised? An Empirical Study of Mine Clos-
ings,” Review of Financial Studies, 15(1), 35–64.
Mount, T. D., Y. Ning, and X. Cai, 2006, “Predicting price spikes in electricity markets using a regime-
switching model with time-varying parameters,”Energy Economics, 28, 62–80.
Quigg, L., 1993, “Empirical Testing of Real Option-PricingModels,”The Journal of Finance, 48(2), 621–
640.
Thompson, M., M. Davison, and H. Rasmussen, 2004, “Valuation and Optimal Operation of Electric Power
Plants in Competitive Markets,”Operations Research, 52(4), 546–562.
Tseng, C.-L., and G. Barz, 2002, “Short-Term Generation Asset Valuation: A Real Options Approach,”
Operations Research, 50(2), 297–310.
Ullrich, C. J., 2008, “Equilibrium forward premia in electricity markets,” working paper, Virginia Poly-
technic Institute and State University.
23
Table IPlant Summary Statistics
The table presents summary statistics for the age (to the nearest year), size (megawatts, MW), andefficiency (%) of plants in the sample. The ages are calculated based upon the first year a plant appears inthe sample. SeeSection III: Data, Plant Data for the definition ofRegulated.
Age (yrs) Size (MW) EfficiencyRegulated NOBS 488 488 488
Mean 17.7 49.7 25.2%Stdev 14.2 43.1 4.4%Min 0 0.4 9.2%Max 40 183.0 35.2%
Non-Regulated NOBS 633 633 633Mean 19.3 38.0 24.3%Stdev 14.0 38.5 4.8%Min 0 0.6 5.4%Max 60 246.0 41.8%
Total NOBS 1,121 1,121 1,121Mean 18.6 43.1 24.7%Stdev 14.1 41.0 4.6%Min 0 0.4 5.4%Max 60 246.0 41.8%
Table IIMacro and Real Options Variables Summary Statistics
The table presents summary statistics for macroeconomic and real options variables.REGUNCERTis anindicator variable which takes the value of one during periods of regulatory uncertainty. See thediscussion in Section III for details.
Macro Real OptionsReserve Margin Interest RateSpread Vol ($/MWh) REGUNCERT
NOBS 8,189 8,189 8,189 8,189Mean 19.2% 4.70% $31.19 0.205Stdev 6.2% 0.58% $15.23 0.404Min 11.5% 4.01% $12.07 0Max 30.1% 6.03% $187.44 1
24
Table IIIShutdown: Transitions from OP to OP/SB by Year
Number of plants classified as operating (OP) in thefrom yearand either operating (OP) or shutdown(SB) in theto year.
from year to year OP SB Total2001 2002 695 2 6972002 2003 803 1 8042003 2004 808 43 8512004 2005 820 12 8322005 2006 829 16 8452006 2007 848 0 8482007 2008 851 2 8532008 2009 885 0 885
Total 6,539 76 6,615
Table IVShutdown: Univariate Statistics
Conditional on a plant operating in yeart, the table presents statistics for macroeconomic variables,plant-specific variables, and real options variables (i.e., measures of uncertainty) for plants whichcontinued to operate (did not shutdown, OP) in yeart +1 and those which shutdown (SB) in yeart +1.Regulatory Status is an indicator variable which equals zero if the plant is non-regulated and one if theplant is regulated.
Type Variable OP SB deltaMacro Reserve Margin (%) 19.1% 26.9% -7.8%∗∗∗
Interest Rate (%) 4.68% 4.49% 0.19%∗∗∗
Plant Regulatory Status 0.544 0.224 0.321∗∗∗
Age (years) 21.4 24.4 -3.1∗∗
Efficiency (%) 24.8% 23.4% 1.4%∗∗
Size (MW) 45.1 31.9 13.3∗∗∗
Real Options Spark Spread Stdev ($/MWh) $31.04 $21.37 $9.66∗∗∗
Regulatory Uncertainty Dummy 0.240 0.013 0.227∗∗∗
25
Table V: Shutdown Logit Estimation Results
Consider planti which burns fuelj and is located in regionk. The full model is given by
ISBi,t+1 = α+(β1∗RMk,t+1)+ (β2∗T10t)+ (β3∗REGSTi)+ (β4∗EFFi)+ (β5∗SIZEi)
+(β6∗SPRDSDi jk,t)+ (β7∗REGUNCERTt)+ ε.
The dependent variableISBi,t+1 is an indicator which is equal to zero if planti was operating both in yeart and in yeart +1, and equal to one if
plant i was operating in yeart and shutdown in yeart +1. RMk,t+1 is the projected reserve margin for regionk for yeart +1. T10t is the ten yeartreasury bond rate for yeart. REGSTi is an indicator variable equal to zero if the plant is non-regulated and equal to one if the plant is regulated.EFFi is the efficiency of planti. SIZEi is the capacity of planti. SPRDSDi jk,t is the standard deviation of yeart spark spread for planti whichburns fuel j and is located in regionk. REGUNCERTt is the yeart retail competition index. The table presents the average marginal effects(
∂Prob(ISB= 1)/∂x)
of each independent (x) variable. For the indicator variables (REGSTandREGUNCERT) the table presents the change inthe probability of a shutdown when the variable changes fromzero to one.∗∗∗indicates significance at the 1% level,∗∗indicates significance at the5% level, and∗indicates significance at the 10% level. Each regression has6,515 observations.
RM 0.252∗∗∗ 0.224∗∗∗
T10 -0.902∗∗∗ 0.824∗∗
REGST -0.015∗∗∗ -0.010∗∗∗
EFF -0.064∗∗ -0.045SIZE -0.133∗∗ -0.071SPRDSD -1.016∗∗∗ -0.600∗
REGUNCERT -0.014∗∗∗ -0.012∗∗∗
pseudo-R2 14.3% 1.2% 3.9% 0.7% 1.3% 6.0% 4.0% 22.1%Log-likelihood -355.8 -409.9 -398.9 -412.0 -409.8 -390.3 -398.4 -323.3AIC 715.6 823.8 801.7 828.0 823.7 784.5 800.9 662.6BIC 729.2 837.4 815.3 841.6 837.2 798.1 814.5 717.0
26
Table VIStartup and Abandonment: Transitions from SB to OP/SB/RE byYear
Number of plants classified as shutdown (SB) in thefrom yearand either operating (OP), shutdown (SB),or retired (RE) in theto year.
from year to year OP SB RE Total2001 2002 60 221 1 2822002 2003 47 198 1 2462003 2004 9 143 49 2012004 2005 22 153 13 1882005 2006 1 158 6 1652006 2007 6 173 0 1792007 2008 32 139 2 1732008 2009 7 127 6 140
Total 184 1,312 78 1,574
Table VIIStartup: Univariate Statistics
Conditional on a plant being shutdown in yeart, the table presents statistics for macroeconomic variables,plant-specific variables, and real options variables (i.e., measures of uncertainty) for plants whichremained shutdown (SB) in yeart +1 and which started up (moved to operating, OP) in yeart +1.Regulatory Status is an indicator variable which equals zero if the plant is non-regulated and one if theplant is regulated.
Type Variable SB OP deltaMacro Reserve Margin (%) 18.8% 16.4% 2.4%∗∗∗
Interest Rate (%) 4.78% 5.13% -0.35%∗∗∗
Plant Regulatory Status 0.130 0.103 0.027Age (years) 23.8 21.9 1.9∗
Efficiency (%) 23.2% 24.2% -1.0%∗∗∗
Size (MW) 31.6 46.6 -15.0∗∗∗
Time Shutdown (years) 2.55 1.16 1.39∗∗∗
Real Options Spark Spread Stdev ($/MWh) $32.27 $36.10 -$3.83∗∗∗
Regulatory Uncertainty Dummy 0.075 0.043 0.031∗
27
Table VIII: Startup Logit Estimation Results
Consider planti which burns fuelj and is located in regionk. The full model is given by
IOPi,t+1 = α+(β1∗RMk,t+1)+ (β2∗T10t)+ (β3∗REGSTi)+ (β4∗EFFi)+ (β5∗SIZEi)
+(β6∗SBTIMEi,t)+ (β7∗SPRDSDi jk,t)+ (β8∗REGUNCERTt)+ ε.
The dependent variableIOPi,t+1 is an indicator which is equal to zero if planti was shutdown both in yeart and in yeart +1, and equal to one if
plant i was shutdown in yeart and operating in yeart +1. RMk,t+1 is the projected reserve margin for regionk for yeart +1. T10t is the ten yeartreasury bond rate for yeart. REGSTi is an indicator variable equal to zero if the plant is regulated and equal to one if the plant is non-regulated.EFFi is the efficiency of planti. SIZEi is the capacity of planti. SBTIMEi,t is the length of time, in years, that planti has been shutdown as ofyeart. SPRDSDi jk,t is the standard deviation of yeart spark spread for planti which burns fuelj and is located in regionk. REGUNCERTt is theyeart retail competition index. The table presents the average marginal effects
(
∂Prob(ISB= 1)/∂x)
of each independent (x) variable. For theindicator variables (REGSTandREGUNCERT) the table presents the change in the probability of a startup when the variable changes from zeroto one.∗∗∗indicates significance at the 1% level,∗∗indicates significance at the 5% level, and∗indicates significance at the 10% level. Eachregression has 1,496 observations.
RM -0.814∗∗∗ -0.778∗∗∗
T10 7.615∗∗∗ -2.833∗
REGST -0.026 -0.035EFF 0.483∗∗ 0.242SIZE 0.889∗∗∗ 0.873∗∗∗
SBTIME -0.040∗∗∗ -0.036∗∗∗
SPRDSD 1.435∗∗∗ 0.440REGUNCERT -0.051∗ -0.052∗
pseudo-R2 2.5% 3.8% 0.1% 0.6% 1.9% 6.8% 0.8% 0.2% 11.2%Log-likelihood -543.7 -536.9 -557.2 -554.3 -547.0 -519.8 -553.5 -556.4 -495.4AIC 1,091 1,078 1,118 1,113 1,098 1,044 1,111 1,117 1,009BIC 1,102 1,088 1,129 1,123 1,109 1,054 1,122 1,127 1,057
28
Table IXAbandonment: Univariate Statistics
Conditional on a plant being shutdown in yeart, the table presents statistics for macroeconomic variables,plant-specific variables, and real options variables (i.e., measures of uncertainty) for plants whichremained shutdown (SB) in yeart +1 and those which were abandoned (retired, RE) in yeart +1.Regulatory Status is an indicator variable which equals zero if the plant is non-regulated and one if theplant is regulated.
Type Variable SB RE deltaMacro Reserve Margin (%) 18.8% 27.0% -8.2%∗∗∗
Interest Rate (%) 4.78% 4.51% 0.27%∗∗∗
Plant Regulatory Status 0.130 0.051 0.079∗∗∗
Age (years) 23.8 31.0 -7.2∗∗∗
Efficiency (%) 23.2% 20.7% 2.5%∗∗∗
Size (MW) 31.6 11.9 19.8∗∗∗
Time Shutdown (years) 2.55 2.55 0.00Real Options Spark Spread Stdev ($/MWh) $32.27 $23.39 $8.88∗∗∗
Regulatory Uncertainty Dummy 0.075 0.026 0.049∗∗
29
Table X: Abandonment Logit Estimation Results
Consider planti which burns fuelj and is located in regionk. The full model is given by
IREi,t+1 = α+(β1∗RMk,t+1)+ (β2∗T10t)+ (β3∗REGSTi)+ (β4∗EFFi)+ (β5∗SIZEi)
+(β6∗SBTIMEi,t)+ (β7∗SPRDSDi jk,t)+ (β8∗REGUNCERTt)+ ε.
The dependent variableIREi,t+1 is an indicator which is equal to zero if planti was shutdown both in yeart and in yeart +1, and equal to one if
plant i was shutdown in yeart and abandoned in yeart +1. RMk,t+1 is the projected reserve margin for regionk for yeart +1. T10t is the tenyear treasury bond rate for yeart. REGSTi is an indicator variable equal to zero if the plant is regulated and equal to one if the plant isnon-regulated.EFFi is the efficiency of planti. SIZEi is the capacity of planti. SBTIMEi,t is the length of time, in years, that planti has beenshutdown as of yeart. SPRDSDi jk,t is the standard deviation of yeart spark spread for planti which burns fuelj and is located in regionk.REGUNCERTt is the yeart retail competition index. The table presents the average marginal effects
(
∂Prob(ISB= 1)/∂x)
of each independent(x) variable. For the indicator variables (REGSTandREGUNCERT) the table presents the change in the probability of an abandonment whenthe variable changes from zero to one.∗∗∗indicates significance at the 1% level,∗∗indicates significance at the 5% level, and∗indicatessignificance at the 10% level. Each regression has 1,390 observations.
RM 1.184∗∗∗ 1.336∗∗∗
T10 -4.626∗∗∗ 6.742∗∗∗
REGST -0.038∗∗∗ 0.018EFF -0.632∗∗∗ -0.859∗∗∗
SIZE -2.760∗∗∗ -2.540∗∗∗
SBTIME -0.000 0.026∗∗∗
SPRDSD -3.549∗∗∗ -1.695∗∗
REGUNCERT -0.039∗∗∗ -0.026pseudo-R2 22.2% 2.6% 0.9% 3.8% 7.9% 0.0% 5.9% 0.6% 37.5%Log-likelihood -233.9 -292.6 -297.9 -289.0 -276.8 -300.4 -282.7 -298.7 -187.9AIC 471.8 589.3 599.7 582.1 557.5 604.9 569.4 601.5 393.8BIC 482.3 599.8 610.2 592.6 568.0 615.3 579.9 611.9 440.9
30
Table XI: Startup and Abandonment Multinomial Logit Estimation Results
Consider planti which burns fuelj and is located in regionk. The full model is given by
IOPREi,t+1 = α+(β1∗RMk,t+1)+ (β2∗T10t)+ (β3∗REGSTi)+ (β4∗EFFi)+ (β5∗SIZEi)
+(β6∗SBTIMEi,t)+ (β7∗SPRDSDi jk,t)+ (β8∗REGUNCERTt)+ ε.
The dependent variableIOPREi,t+1 is an indicator which is equal to zero if planti was on standby in yeart and operating in yeart +1, equal to one if
plant i was on standby both in yeart and in yeart +1, equal to two if planti was on standby in yeart and retired in yeart +1. RMk,t+1 is theprojected reserve margin for regionk for yeart +1. T10t is the ten year treasury bond rate for yeart. REGSTi is an indicator variable equal tozero if the plant is regulated and equal to one if the plant is non-regulated.EFFi is the efficiency of planti. SIZEi is the capacity of planti.SBTIMEi,t is the length of time, in years, that planti has been shutdown as of yeart. SPRDSDi jk,t is the standard deviation of yeart spark spreadfor plant i which burns fuelj and is located in regionk. REGUNCERTt is the yeart retail competition index. The table presents the averagemarginal effects
(
∂Prob(ISB= 1)/∂x)
of each independent (x) variable. For the indicator variables (REGSTandREGUNCERT) the tablepresents the change in the probability when the variable changes from zero to one.∗∗∗indicates significance at the 1% level,∗∗indicatessignificance at the 5% level, and∗indicates significance at the 10% level. Each regression has1,574 observations.
Startup (SB to OP) Abandon (SB to RE)RM -0.794∗∗∗ 1.203∗∗∗
T10 -2.922∗∗ 5.878∗∗∗
REGST -0.035∗ 0.017EFF 0.260 -0.755∗∗∗
SIZE 0.935∗∗∗ -2.247∗∗∗
SBTIME -0.035∗∗∗ 0.024∗∗∗
SPRDSD 0.486 -1.468∗∗
REGUNCERT -0.048 -0.024pseudo-R2 21.2%Log-likelihood -684.0AIC 1,404BIC 1,500
31
Figure 1Shutdown ProbabilityThe top panel presents the probability of shutting down an operating plant as a function for reserve margin forthe cases of regulatory uncertainty (blue circles) and no uncertainty (red squares). The bottom panel presents theprobability of shutting down an operating plant as a function for reserve margin for three values of spark spreadvolatility - $10/MWh (blue circles), $30/MWh (red squares), and $100/MWh (green triangles).
0.00
0.02
0.04
0.06
0.08
0.10
Pr(
Shu
tdow
n)
0.10 0.15 0.20 0.25 0.30Reserve Margin
Regulatory Uncertainty No Uncertainty
0.00
0.02
0.04
0.06
0.08
0.10
Pr(
Shu
tdow
n)
0.10 0.15 0.20 0.25 0.30Reserve Margin
$10/MWh $30/MWh $100/MWhSpark Spread Volatility
32
Figure 2Startup ProbabilityThe top panel presents the probability of starting up a plantwhich was previously shutdown as a function for reservemargin for the cases of regulatory uncertainty (blue circles) and no uncertainty (red squares). The bottom panelpresents the probability of starting up a plant which was previously shutdown as a function for reserve margin forthree values of spark spread volatility - $10/MWh (blue circles), $30/MWh (red squares), and $100/MWh (greentriangles).
0.00
0.05
0.10
0.15
0.20
0.25
Pr(
Sta
rtup
)
0.10 0.15 0.20 0.25 0.30Reserve Margin
Regulatory Uncertainty No Uncertainty
0.00
0.05
0.10
0.15
0.20
0.25
Pr(
Sta
rtup
)
0.10 0.15 0.20 0.25 0.30Reserve Margin
$10/MWh $30/MWh $100/MWhSpark Spread Volatility
33
Figure 3Abandonment ProbabilityThe top panel presents the probability of abandoning a plantwhich was previously shutdown as a function for reservemargin for the cases of regulatory uncertainty (blue circles) and no uncertainty (red squares). The bottom panelpresents the probability of abandoning a plant which was previously shutdown as a function for reserve margin forthree values of spark spread volatility - $10/MWh (blue circles), $30/MWh (red squares), and $100/MWh (greentriangles).
0.00
0.10
0.20
0.30
0.40
0.50
Pr(
Aba
ndon
)
0.10 0.15 0.20 0.25 0.30Reserve Margin
Regulatory Uncertainty No Uncertainty
0.00
0.10
0.20
0.30
0.40
0.50
Pr(
Aba
ndon
)
0.10 0.15 0.20 0.25 0.30Reserve Margin
$10/MWh $30/MWh $100/MWhSpark Spread Volatility
34
Figure 4Startup and Abandonment ProbabilityFor plants which were previously shutdown, the figure present the probability of startup (OP, red squares), remainingon standby (SB, blue circles), and abandonment (RE, green triangles) as a function of reserve margin. The toppanel shows the probabilities for no regulatory uncertainty (left) and regulatory uncertainty (right). The bottompanel shows the probabilities for low spark spread volatility of ($10/MWh, left) and high spark spread volatility($100/MWh right).
0.00
0.20
0.40
0.60
0.80
1.00
Pro
babi
lty
0.10 0.15 0.20 0.25 0.30Reserve Margin
OP SB RE
No Uncertainty
0.00
0.20
0.40
0.60
0.80
1.00
Pro
babi
lty
0.10 0.15 0.20 0.25 0.30Reserve Margin
OP SB RE
Regulatory Uncertainty
0.00
0.20
0.40
0.60
0.80
1.00
Pro
babi
lty
0.10 0.15 0.20 0.25 0.30Reserve Margin
OP SB RE
Spread Vol = $10/MWh
0.00
0.20
0.40
0.60
0.80
1.00
Pro
babi
lty
0.10 0.15 0.20 0.25 0.30Reserve Margin
OP SB RE
Spread Vol = $100/MWh
35
Figure 5Histogram of Capacity for Startup and AbandonmentFor plants which were previously shutdown, the figure present histograms of capacity. The top panel shows thedistribution of capacity for plants which startup. The bottom shows the distribution of capacity for plants which areabandoned.
0.1
.2.3
.4.5
.6F
ract
ion
0 25 50 75 100 125 150Capacity (MW)
Startup
0.1
.2.3
.4.5
.6F
ract
ion
0 25 50 75 100 125 150Capacity (MW)
Abandon
36
Figure 6Startup and Abandonment Probability for Large and Small PlantsFor plants which were previously shutdown, the figure present the probability of starting up (red squares), remainingshutdown (blue circles), and abandoning (green triangles)as a function of reserve margin for spark spread volatilitiesof $10/MWh (left) and $100/MWh (right). The top panel shows the probabilities for large plants, greater than 25MW. The bottom panel shows the probabilities for small plants, less than 25 MW.
0.00
0.20
0.40
0.60
0.80
1.00
Pro
babi
lty
0.10 0.15 0.20 0.25 0.30Reserve Margin
OP SB RE
Spread Vol = $10/MWh
0.00
0.20
0.40
0.60
0.80
1.00
Pro
babi
lty
0.10 0.15 0.20 0.25 0.30Reserve Margin
OP SB RE
Spread Vol = $100/MWh
Large Plants (>25 MW)
0.00
0.20
0.40
0.60
0.80
1.00
Pro
babi
lty
0.10 0.15 0.20 0.25 0.30Reserve Margin
OP SB RE
Spread Vol = $10/MWh
0.00
0.20
0.40
0.60
0.80
1.00
Pro
babi
lty
0.10 0.15 0.20 0.25 0.30Reserve Margin
OP SB RE
Spread Vol = $100/MWh
Small Plants (<25 MW)
37
Appendix A. Status Codes SB and BU - Definitions and Changes
For every year, EIA provides variable definitions in aLayoutfile accompanying the EIA 860 data. Status
code SB is not defined in theLayout file for the 2001 and 2002 years. However, the 2000Layout file
defines SB as
“Cold Standby (Reserve): deactivated (mothballed), in long-term storage and cannot be made available
for service in a short period of time, usually requires threeto six months to reactivate.”
Beginning in 2003 SB is defined as
“Standby - available for service but not normally used (has little or no generation during the year).”
Status code BU is available only for the 2004-2006 time period. For the 2004-2006 time period, the
definition of SB is unchanged. BU is defined as
“Backup - used for test purposes or emergency such as shortage to power to meet load requirements.”
For the 2007-2009 time period, BU again disappears and SB is defined as
“Standby/Backup - available for service but not normally used (has little or no generation during the
year).”
38
Appendix B. Heat Rate Data
Heat rate data is available for 943 of the 1,121 plants in our sample. In order to estimate the heat rates
of the remaining 178 plants, we calculate mean heat rates by size and in-service year. The heat rate for a
combustion turbine varies (1) inversely with the size of theplant (bigger machines are more efficient), and
(2) directly with the age of the plant (newer machines are much more efficient). We classify plants into size
and age categories (five of each) and then calculate the average heat rate in each age-size category based
upon the heat rates available from CEMS and Form 860. We then use these average heat rates for other
plants in these size-age categories.
For example, heat rate data is available for 318 plants whichwent into service in the 1970s and with
capacity less than 50MW. The average heat rate for these 318 plants is 16.055 MMBtu/MWh. There are 16
plants which fall into the same size-age category and for which no heat rate data are available. For those
16 plants we assign the heat rate to be 16.055 MMBtu/MWh.
Heat rate data is available for 26 plants which went into service in the 2000s and with capacity in the
100-150 MW range. The mean heat rate for these 26 plants is 11.880 MMBtu/MWh. There are 5 plants
which fall into the same size-age category and for which no heat rate data are available. For those 5 plants
we assign the heat rate to be 11.880 MMBtu/MWh. And so forth and so on.
39
Appendix C. Startup and Shutdown Costs
Most of the problems encountered in restarting a plant are associated with the control system, i.e., instru-
mentation, electronic controls, and wiring. In general these systems do not vary greatly with the size of
the plant in question. Mechanical issues involved in shutdown and restart are primarily concerned with
corrosion. Core preservation requires layup chemicals.18
Restarting a plant begins with checking the control loops. Maintenance personnel attempt to “shoot-
the-loop”, i.e., to check that each control loop is functioning and, if not, to determine where the problem
lies. It is common for systems that were in perfect working order at the time the plant was shutdown to fail
when restart is attempted.
The costs to restart a plant also can vary with the corporate culture of the owner. Oftentimes mainte-
nance of shutdown plants has a lower priority than maintaining operating plants. A willingness to spend
money to maintain these systems while the plant is shutdown greatly reduces the one time cost associated
with the actual restart. However, management may not perceive that spending money on a plant which is
not currently operating is a wise investment.
The unfortunate (for our purposes) conclusion is that two plants which are the same size, same age, and
located in the same region can have very different shutdown and startup costs depending on the priorities
of the management team.
In summary, there is no simple way to estimate the costs associated with shutting down and restarting
a plant based strictly upon the data available from EIA. Eachplant is unique and each firm is unique.
As discussed in the main text, we focus on simple cycle gas turbines only, thereby eliminating variation
across technology types. The control system issues discussed above should not vary much with the capacity
of the plant.
18For example, the introduction of nitrogen can prevent oxygen from coming into contact with the core and causing corrosion.
40